Hierarchical Exploration and Visualization of Large and Complex Multi-Dimensional Data Sets

这个项目研究了使用分层方法来访问和可视化大型和复杂的多维数据集。可视化可以提供一种有效而强大的方法来探索这些数据。然而，大多数可视化方法是为中等大小的非分层数据集开发的，这些数据集可以在图形工作站上以可接受的速度成像。在实践中，有许多数据集不能使用这些方法有效地探索。该项目正在开发和实现为各种网格类型上的非常大的多维数据集设计的可视化方法。这些数据集在时间和空间资源管理方面存在重大问题，本研究将解决这些问题。统一的主题将是使用分层数据结构来有效地管理这两种资源。为了解决时间效率问题，正在开发用于选择性遍历和数据快速交互式成像的技术。为了提高空间效率，研究了各种数据模型，如小波，以实现压缩，同时可以直接修改为可视化。这些层次结构与现有的可视化技术集成在一起，用于等值面提取、直接体绘制和流动可视化，其中许多技术已经在加州大学圣克鲁斯分校和其他地方开发出来，用于非层次数据。研究了直线网格、结构化网格和非结构化网格的技术。SIMD和MIMD机器的这些方法的并行化也正在探索中。这项研究正在为大型复杂数据集的可视化开发新的算法，并使软件系统可供科学家使用。软件是模块化的，以便与现有的可视化软件包集成。虽然主要目标是可视化，但所开发的分层方法也将用于处理非常大的数据集时因访问问题而受到阻碍的任何数据分析。

This project investigates the use of hierarchical methods for accessing and visualizing large and complex multi- dimensional data sets. Visualization can provide an efficient and powerful means of exploring such data However, most visualization methods are developed for moderately sized non-hierarchial data sets that can be imaged acceptably fast on graphics workstations. In practice, there are many data sets that cannot be effectively explored using these methods. The project is developing and implementing visualization approches designed for very large, multi-dimensional data sets over a variety of grid types. Such data sets present significant problems in the management of time and space resources, which will be addressed by this research. The unifying theme will be the use of hierarchical data structures to manage both resources effectively. Addressing time efficiency, techniques are being developed for the selective traversal and rapid interactive imaging of the data. For space efficiency, various data models, such as wavelets, are investigated to achieve compression, while being directly amendable to visualization. These hierarchies are being integrated with existing visualization techniques for isosurface extraction, direct volume rendering, and flow visualization, many of which have already been developed at UC Santa Cruz and elsewhere for non-hierarchial data. Techniques for rectilinear, structured, and unstructured grids are studied. Parallelization of these methods of SIMD and MIMD machines is also is being explored. This research is developing new algorithms for visualization of large, complex data sets, as well as making software systems available for use by scientists. Software is modular to facilitate integration with existing visualization packages. While the main objective is visualization, the hierarchical methods developed will also be of use for any data analysis hampered by access problems in dealing with very large data sets.